The Agent Skills Directory

[SAFE]: The skill is designed for academic text refinement with clear boundaries and deterministic quality controls. It does not perform network operations, access sensitive files, or execute untrusted remote code.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted text from user manuscripts which represents a potential injection surface. However, the skill explicitly mitigates this risk through anti-hallucination instructions and the mandatory use of scripts/check_rewrite_fidelity.py, which ensures that technical data and citations remain unchanged regardless of the agent's rewriting process.
Ingestion points: manuscript and section text provided via the prompt or files.
Boundary markers: Not explicitly defined in the prompt interpolation, but the workflow is strictly phased.
Capability inventory: File system access (Read, Write, Edit), search tools (Grep, Glob), and shell execution (Bash) for internal scripts.
Sanitization: Relies on post-execution fidelity verification rather than input filtering.
[COMMAND_EXECUTION]: The skill utilizes the Bash tool to run local Python scripts (check_rewrite_fidelity.py, check_sentence_variety.py) included in the skill package. These scripts are implemented using standard libraries and focus on character/word count analysis and regex-based pattern matching. No arbitrary command execution or shell injection vulnerabilities were detected in the script logic.
[EXTERNAL_DOWNLOADS]: The skill documentation references external sources for provenance (e.g., academic repositories and local rule paths like ~/.claude/rules/), but it does not perform any runtime downloads or remote code execution. All logic is contained within the provided scripts and markdown references.

humanize